On 10.12.2025 06:23, Khushit Shah wrote:
Hi,

I wanted to follow up on this patch to see if there are any review comments
or feedback. I'm planning to prepare a v2 that addresses the following:

1. Move the SEOIB configuration code from x86-common.c to KVM-specific code
    (kvm_arch_init()).

2. Refactor as per the changes on the KVM side of the patch.

Before proceeding with v2, I have a design question regarding the scope of
the fix:

Currently, the patch sets the SEOIB state for all machine types on new power-ons
based on the IOAPIC version. This means that any new VM powered on with a
patched QEMU will get the proper SEOIB behavior.

However, I'm wondering if we should instead:
- Define a new machine property (i.e, "seoib-policy") that defines the SEOIB
   behavior.
- Only enable the new SEOIB behavior in the latest machine type version
   (10.2?), keeping older machine types in QUIRKED mode.

The question is: should new power-ons of *all* machine types set the SEOIB
state automatically, or should we scope this fix to the latest machine type
versions only via a machine property?

I'd appreciate your thoughts on this design decision before I finalise v2.

Now that the KVM side of this patch is in Linus tree are there any plans
to submit a v2 for visibility and with updated flag naming even before
getting the questions above answered?

Regards,
Khushit


Thanks,
Maciej


Reply via email to